The release archives that the Woosh-DFlow Space unpacks into its image, mirrored from the v1.0.0 release of Woosh by Sony AI. The archives are unchanged, each extracting to its own checkpoints/<name> folder.
Woosh-DFlow.zip: the distilled latent diffusion modelWoosh-AE.zip: the audio autoencoderTextConditionerA.zip: the audio-CLAP text conditioner
The weights are licensed CC BY-NC 4.0, so they may be used non-commercially only, with attribution to Sony AI.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support